Mask set inspection device

Abstract

The present invention provides a mask set inspection device that can accurately inspect the presence or absence of foreign matter on the mask, flat glass, and pellicle individually without separating them. [Solution] The mask set inspection device 1 includes an illumination device 3 that irradiates light onto the mask set 200, a CIS 4 having a lens array 41 in which cylindrical lenses are arranged in parallel and a photoelectric conversion element array 42 in which a plurality of photoelectric conversion elements are arranged in a row to form an image when light passes through the cylindrical lenses, and a focus position adjustment mechanism 6 that can adjust the focus position of the CIS 4 to match the mask 202, the flat glass 201 and the pellicle 203, and determines the presence or absence of foreign matter based on image data obtained by the CIS 4. The CIS 4 has a focal length that is greater than the distance from the top surface of the flat glass 201 to the top surface of the pellicle 203, but less than 60 mm.

Description

【Technical Field】 【0001】 The present invention relates to a mask set inspection apparatus for inspecting foreign matter adhesion to a mask set having a mask, a flat glass, and a pellicle arranged so as to sandwich the mask with a predetermined interval therebetween. 【Background Art】 【0002】 Semiconductor devices, liquid crystal display devices, etc. are manufactured through a process of transferring a circuit pattern formed on a mask onto a photosensitive substrate. 【0003】 By the way, in order to prevent problems caused by using a mask alone, a mask set in which the mask is sandwiched between a flat glass and a pellicle may be used. The flat glass is provided on the side opposite to the circuit pattern surface of the mask with a predetermined interval therebetween, and forms a sealed chamber with the mask. When the mask is bent due to its own weight, the bending is corrected by adjusting the pressure in the sealed chamber. Also, the pellicle is provided on the circuit pattern surface side of the mask with a predetermined interval therebetween, and covers and protects the circuit pattern. 【0004】 When transferring a circuit pattern using the above mask set, if foreign matter such as dust or dirt adheres to the mask set, there is a risk that not only the circuit pattern but also an image of the foreign matter will be transferred onto the photosensitive substrate. Therefore, prior to pattern transfer (exposure), it may be necessary to check for the presence or absence of foreign matter adhesion to the mask set. 【0005】 A mask set inspection device has been proposed for inspecting for the adhesion of foreign matter to a mask set, comprising a first inspection unit located on the flat glass side and a second inspection unit located on the pellicle side (see, for example, Patent Document 1). Each inspection unit comprises a light-emitting unit that projects a predetermined inspection light (illumination light) onto the surface of the mask set to be inspected, and a light-receiving unit that receives scattered light (reflected light) from foreign matter generated by the light-emitting unit, and the presence or absence of foreign matter is inspected based on the light-receiving result by the light-receiving unit. The first inspection unit inspects for the presence or absence of foreign matter to the flat glass, and the second inspection unit inspects for the presence or absence of foreign matter to the pellicle. Furthermore, in this mask set inspection device, the flat glass is moved (retracted) and the height position of the first inspection unit is adjusted so that the first inspection unit can inspect for the presence or absence of foreign matter to the mask. [Prior art documents] [Patent Documents] 【0006】 [Patent Document 1] Japanese Patent Publication No. 2020-51759 [Overview of the Initiative] [Problems that the invention aims to solve] 【0007】 However, the aforementioned mask set inspection device requires two inspection units, each having a light-emitting unit and a light-receiving unit, positioned to sandwich the mask set. In other words, the light-emitting unit and light-receiving unit must be provided on both the flat glass side and the pellicle side. This could lead to increased manufacturing and maintenance costs for the device, as well as the device becoming larger and more complex. 【0008】 Furthermore, in order to inspect the masks, it is necessary to move the flat glass, but moving the flat glass may cause contamination of the masks or other items during the process. 【0009】 Furthermore, because the inspection is based on the reception results of scattered light (reflected light) from foreign objects, there is a risk that foreign objects that affect the transfer (exposure) of circuit patterns may not be properly detected. 【0010】 The present invention has been made in view of the above circumstances, and its purpose is to provide a mask set inspection device that can accurately inspect for the presence or absence of foreign matter on the mask, flat glass, and pellicle individually without separating them, while suppressing cost increases and miniaturizing and simplifying the device, and furthermore, can more appropriately detect foreign matter that affects the transfer (exposure) of the circuit pattern. [Means for solving the problem] 【0011】 Below, we will describe, in separate sections, each means suitable for achieving the above objectives. Furthermore, we will add notes on the effects and benefits specific to each means as needed. 【0012】 Means 1. A mask set inspection apparatus for inspecting for the adhesion of foreign matter to a mask set having a mask on which a circuit pattern for transfer onto a substrate is formed, and a transparent flat glass and a pellicle, respectively, arranged so as to sandwich the mask at a predetermined distance from the mask, An irradiation means for irradiating the mask set with a predetermined light, The system includes a lens array formed by arranging multiple cylindrical lenses in parallel, and a photoelectric element array formed by arranging multiple photoelectric elements that image light passing through the cylindrical lenses, and an imaging means provided at a position sandwiching the mask set between the irradiation means and the mask set, capable of imaging light irradiated by the irradiation means and transmitted through the mask set. The mask, the flat glass, and the pellicle are each to be inspected, and the focusing position adjustment means is capable of adjusting the focusing position of the imaging means to match the inspected object. The system includes a determination means capable of determining the presence or absence of foreign matter in the object to be inspected based on image data obtained by the imaging means, The mask set inspection apparatus is characterized in that the imaging means has a focal length greater than the distance from the surface of the flat glass on the side of the imaging means to the surface of the pellicle on the side of the imaging means, but less than 60 mm. 【0013】 Furthermore, focal length refers to the distance from the lens array that forms an image on the photoelectric conversion element (i.e., is in a conjugate relationship) to the object (the target of imaging). 【0014】 Generally, the depth of field increases with increasing shooting distance. However, according to the above-described means 1, the imaging means has a focal length of less than 60 mm, resulting in a short shooting distance and a shallow depth of field. Therefore, when the imaging means captures an image of the target inspection object (mask, flat glass, or pellicle) after adjusting the focus position of the imaging means to match the target inspection object, the resulting image data will show that the circuit pattern of the mask and foreign matter attached to inspection objects other than the target inspection object are blurred, while foreign matter attached to the target inspection object is clearly visible. For example, if the inspection object is a mask, the resulting image data will show that foreign matter attached to the flat glass or pellicle is blurred, while foreign matter attached to the mask is clearly visible. Also, for example, if the inspection object is flat glass, the resulting image data will show that the circuit pattern of the mask and foreign matter attached to the mask or pellicle are blurred, while foreign matter attached to the flat glass is clearly visible. Therefore, the presence or absence of foreign matter on the mask, flat glass, and pellicle can be accurately inspected individually without separating them. Furthermore, since there is no need to move the inspection targets individually during the inspection, contamination of the mask and other components due to movement can be more reliably prevented. 【0015】 Furthermore, while the circuit pattern is transferred onto the substrate by light transmitted through the mask, according to means 1 above, the imaging means captures the light that is irradiated by the illumination means and transmitted through the mask set. Therefore, the imaging means can obtain image data that simulates the transfer of the circuit pattern (image data showing what is transferred onto the substrate). This makes it possible to more appropriately detect foreign objects that affect the transfer (exposure) of the circuit pattern. 【0016】 Furthermore, the imaging means has a focal length greater than the distance from the surface of the flat glass on which the imaging means is located to the surface of the pellicle on which the imaging means is located. Therefore, when adjusting the focus position of the imaging means to match the object being inspected, it is possible to focus on the object being inspected without bringing the imaging means into contact with the flat glass or the like. This allows for accurate inspection of the presence or absence of foreign matter on each object being inspected while preventing damage to the flat glass or the like. 【0017】 Furthermore, according to the above-described means 1, for example, the irradiation means can be placed on the pellicle side and the imaging means on the flat glass side, and it is not necessary to provide the irradiation means and imaging means on both the flat glass side and the pellicle side. Therefore, it is possible to suppress increases in manufacturing and maintenance costs of the device and to miniaturize and simplify the device. 【0018】 Means 2. The mask set inspection apparatus according to Means 1, characterized in that the imaging means has a focal length of less than 40 mm. 【0019】 According to method 2 described above, the focal length of the imaging means is set to less than 40 mm. Therefore, in the image data obtained by the imaging means, the circuit pattern of the mask and foreign matter attached to inspection targets other than the target inspection target become more blurred, while foreign matter attached to the target inspection target becomes more clearly visible. Therefore, the presence or absence of foreign matter can be inspected with greater accuracy. 【0020】 Means 3. Only one imaging means is provided. For each inspection target, after adjusting the focus position of the imaging means by the focus position adjustment means, imaging is performed by the imaging means, and based on the image data obtained by this imaging, the determination means determines the presence or absence of foreign matter. The mask set inspection apparatus according to means 1, characterized in that it is configured as such. 【0021】 According to the above means 3, only one imaging means is provided, and image data related to the mask, image data related to the flat glass, and image data related to the pellicle can be obtained by this one imaging means. Therefore, it is possible to more effectively suppress an increase in cost and to miniaturize and simplify the apparatus. 【0022】 Means 4. The irradiation means A light source that emits ultraviolet light with a wavelength of 365 nm, A diffuser plate that diffuses the ultraviolet light emitted from the light source and irradiates the mask set, and the mask set inspection apparatus according to means 1, characterized in that it has the same. 【0023】 According to the above means 4, the mask set is irradiated with diffused ultraviolet light having a wavelength of 365 nm by the irradiation means. Therefore, it is possible to irradiate the mask set with light similar to the light (so-called i-line) used at the time of circuit pattern transfer (exposure) by the exposure apparatus. As a result, it is possible to more reliably obtain image data (image data indicating what is transferred to the substrate) assuming the time of circuit pattern transfer. As a result, it is possible to more appropriately detect foreign matter that affects circuit pattern transfer (exposure). 【0024】 Means 5. The imaging means is configured such that light passing through a plurality of the cylindrical lenses forms an image on one of the photoelectric conversion elements. The mask set inspection apparatus according to means 1, characterized in that it is configured as such. 【0025】 According to the above method 5, instead of light passing through one cylindrical lens forming an image on a single photoelectric conversion element, light passing through multiple cylindrical lenses forms an image on a single photoelectric conversion element. Therefore, it is possible to more reliably prevent foreign objects from being hidden by circuit patterns, etc., in the resulting image data. This makes it possible to inspect for the presence or absence of foreign objects with even greater accuracy. 【0026】 Means 6. A neural network having an encoding unit for extracting features from input image data and a decoding unit for reconstructing image data from the features, which is trained using only image data relating to the flat glass without foreign objects as training data to generate a flat glass identification means, A mask identification means generated by training the aforementioned neural network with only image data relating to the mask free of foreign objects as training data, The neural network is trained using only image data of the pellicle free of foreign objects as training data to generate a pellicle identification means, The determination means is By comparing the image data relating to the flat glass obtained by the imaging means with the first reconstructed image data, which is image data reconstructed by inputting the image data into the identification means for flat glass, the presence or absence of foreign matter in the flat glass is determined. By comparing the image data relating to the mask obtained by the imaging means with the second reconstructed image data, which is image data reconstructed by inputting the image data into the mask identification means, the presence or absence of foreign matter in the mask is determined. The mask set inspection apparatus according to means 1, characterized in that it is configured to determine the presence or absence of foreign matter on the pellicle by comparing image data relating to the pellicle obtained by the imaging means with a third reconstructed image data which is image data reconstructed by inputting the image data into the pellicle identification means. 【0027】 Furthermore, the "image data relating to the flat glass free of foreign matter," "image data relating to the mask free of foreign matter," and "image data relating to the pellicle free of foreign matter" used as the "training data" above may be based on image data (actual image data) obtained by actually imaging the flat glass, mask, etc. with an imaging means, or they may be based on virtually generated image data free of foreign matter. Examples of actual image data include image data accumulated from previous inspections and image data relating to mask sets free of foreign matter that have been visually selected by an operator. 【0028】 Furthermore, the term "neural network" includes, for example, a convolutional neural network having multiple convolutional layers. The term "learning" includes, for example, deep learning. The term "discrimination means (generative model)" includes, for example, an autoencoder or a convolutional autoencoder. 【0029】 In addition, since the "identification means for flat glass," "identification means for masks," and "identification means for pellicles" are generated by training only on image data without foreign matter, the reconstructed image data generated when image data containing foreign matter is input to these identification means will be almost identical to the input image data from which the noise portion (the portion corresponding to the foreign matter) has been removed. In other words, when there is foreign matter in the target object to be inspected, the reconstructed image data for the target object to be inspected is a hypothetical image data for the target object to be inspected, assuming that there is no foreign matter. 【0030】 According to the above-described means 6, image data relating to the target inspection object (flat glass, mask, or pellicle) obtained by the imaging means is compared with reconstructed image data obtained by inputting the image data into the identification means, and the presence or absence of foreign matter in the target inspection object is determined based on the comparison result. Since both image data to be compared relate to the same inspection object, foreign matter inspection can be performed more easily and accurately. 【0031】 Furthermore, since the shape and appearance of the mask set (e.g., the shape and position of the circuit pattern) are almost identical in both image data being compared, unlike methods that detect foreign objects by comparing them to a reference image, it is not necessary to set relatively lenient inspection conditions to prevent false detections, and stricter inspection conditions can be set. In addition, the state of the object being inspected (e.g., the placement position, placement angle, and deflection of the mask set) and imaging conditions (e.g., lighting conditions and the field of view of the imaging device) can be matched in both image data being compared. These combined effects allow for more accurate inspection of whether or not foreign objects are present. 【0032】 Means 7. The mask set inspection apparatus according to Means 1, characterized by comprising a total of three imaging means: a first imaging means for imaging the flat glass, a second imaging means for imaging the mask, and a third imaging means for imaging the pellicle. 【0033】 According to the above means 7, since the focus position can be adjusted and imaging can be performed simultaneously in each imaging means, image data related to each inspection target can be acquired more efficiently compared to when each inspection target is imaged by a single imaging means. Therefore, the cycle time for inspection can be effectively shortened. 【0034】 Furthermore, the technical aspects related to each of the above means may be combined as appropriate. For example, at least one of the technical aspects related to means 3 to 7 may be combined with the technical aspects related to means 2. [Brief explanation of the drawing] 【0035】 [Figure 1] This is a schematic diagram illustrating the mask set inspection device and other related equipment. [Figure 2] This is a block diagram showing the functional configuration of a mask set inspection device. [Figure 3] This is a schematic diagram showing the general configuration of a lens array and a photoelectric conversion element array. [Figure 4]This is a schematic diagram illustrating an example of a contact image sensor in which only light passing through a single cylindrical lens is imaged for a single photoelectric conversion element. [Figure 5] This is a schematic diagram illustrating an example of a contact image sensor in which only light passing through a single cylindrical lens is imaged for a single photoelectric conversion element. [Figure 6] This is a schematic diagram illustrating the structure of a neural network. [Figure 7] This is a flowchart showing the learning process of a neural network. [Figure 8] This is a schematic diagram showing an example of Ig1, the source image data used for training. [Figure 9] This is a schematic diagram showing an example of Ig2, the source image data used for training. [Figure 10] This is a schematic diagram showing an example of Ig3 image data used for training. [Figure 11] This is a schematic diagram showing an example of training data G1. [Figure 12] This is a schematic diagram showing an example of training data G2. [Figure 13] This is a schematic diagram showing an example of training data G3. [Figure 14] This is a flowchart showing the inspection process. [Figure 15] This is a schematic diagram showing an example of raw image data Ik1 for inspection that does not contain foreign objects. [Figure 16] This is a schematic diagram showing an example of raw image data Ik1 for inspection containing a foreign object. [Figure 17] This is a schematic diagram showing an example of inspection image data K1 free of foreign objects. [Figure 18] This is a schematic diagram showing an example of inspection image data K1 containing a foreign object. [Figure 19] This is a schematic diagram showing an example of the first reconstructed image data S1. [Figure 20] This is a schematic diagram showing an example of Ik2, the raw image data for inspection that does not contain foreign objects. [Figure 21] This is a schematic diagram showing an example of raw image data Ik2 for inspection containing a foreign object. [Figure 22]This is a schematic diagram showing an example of K2 image data for inspection that does not contain foreign objects. [Figure 23] This is a schematic diagram showing an example of K2 image data for inspection containing a foreign object. [Figure 24] This is a schematic diagram showing an example of the second reconstructed image data S2. [Figure 25] This is a schematic diagram showing an example of Ik3, the raw image data for inspection that does not contain foreign objects. [Figure 26] This is a schematic diagram showing an example of raw image data Ik3 for inspection containing a foreign object. [Figure 27] This is a schematic diagram showing an example of K3, an image data for inspection that does not contain any foreign objects. [Figure 28] This is a schematic diagram showing an example of K3, an image data file used for inspection containing a foreign object. [Figure 29] This is a schematic diagram showing an example of third-reconstruction image data S3. [Figure 30] This is a schematic diagram illustrating a mask set inspection device and other components in another embodiment. [Modes for carrying out the invention] 【0036】 An embodiment will be described below with reference to the drawings. The mask set inspection device is a device for inspecting whether or not foreign matter is attached to a mask set. Before describing the mask set inspection device, the mask set to be inspected will be described. 【0037】 As shown in Figure 1, the mask set 200 comprises a mask (photomask) 202 and a flat glass 201 and a pellicle 203 arranged so as to sandwich the mask 202. 【0038】 The mask 202 is a transparent plate made of a predetermined material such as glass or resin, on which a circuit pattern for transfer onto a substrate (photosensitive substrate) is formed. The transfer of the circuit pattern onto the substrate is performed by irradiating the mask 202 with light (ultraviolet light) and allowing the light that has passed through the mask 202 to reach the substrate (exposure process). 【0039】 The flat glass 201 is made of a transparent glass plate and is placed on the opposite side of the mask 202 from the circuit pattern surface (bottom surface) of the mask 202, at a predetermined distance from it. The flat glass 201 forms a sealed chamber 204 between itself and the mask 202. By adjusting the pressure in the sealed chamber 204, it is possible to correct any deflection of the mask 202 caused by its own weight. 【0040】 The pellicle 203 is made of a transparent thin film and is provided on the circuit pattern side of the mask 202 at a predetermined distance from the mask 202. The pellicle 203 covers the circuit pattern and functions as a protective film to prevent foreign matter from adhering to the circuit pattern surface of the mask 202. 【0041】 Next, the mask set inspection device 1 will be described. As shown in Figures 1 and 2, the mask set inspection device 1 comprises a stage 2, an illumination device 3, a contact image sensor 4 (hereinafter referred to as "CIS4"), a scanning mechanism 5, a focus position adjustment mechanism 6, and a control device 7. In this embodiment, the illumination device 3 constitutes the "illumination means," similarly, the CIS4 constitutes the "imaging means," and the focus position adjustment mechanism 6 constitutes the "focus position adjustment means." 【0042】 Stage 2 supports the mask set 200. Stage 2 holds the mask set 200 with the flat glass 201 positioned above, the pellicle 203 below, and the mask 202 etc. in a horizontal position. 【0043】 The lighting device 3 irradiates the mask set 200 with predetermined light from below the stage 2. The lighting device 3 includes a light source 31 and a diffuser plate 32. 【0044】 The light source 31 emits ultraviolet light with a wavelength of 365 nm toward the mask set 200. Ultraviolet light with a wavelength of 365 nm is what is known as the i-line and is commonly used for transferring (exposing) circuit patterns. 【0045】 The diffuser plate 32 diffuses the ultraviolet light emitted from the light source 31 and irradiates the mask set 200 with it. As a result, the ultraviolet light irradiated from the light source 31 is irradiated almost uniformly over the entire area to be inspected in the mask set 200. 【0046】 Only one CIS4 is installed between the illumination device 3 and the mask set 200, and it images the ultraviolet light that is irradiated from the illumination device 3 and passes through the mask set 200. By imaging the mask set 200 from a position close to the mask set 200, the CIS4 acquires image data with very little distortion. The CIS4 is equipped with a lens array 41 and a photoelectric conversion element array 42. 【0047】 The lens array 41 comprises, for example, a plurality of cylindrical lenses 41a, which are 1:1 imaging lenses, and is configured by arranging these cylindrical lenses 41a in parallel (see Figure 3). In this embodiment, the lens array 41 consists of cylindrical lenses 41a arranged in a single row, but it may consist of multiple rows (for example, three rows). 【0048】 The photoelectric element array 42 is composed of multiple photoelectric elements 42a arranged in a row, each of which light passing through a cylindrical lens 41a is imaged. The photoelectric element array 42 is positioned away from the end of the lens array 41, and is configured such that light passing through multiple (for example, three) cylindrical lenses 41a is imaged for each photoelectric element 42a. In other words, the CIS4 in this embodiment is not such that only light passing through one cylindrical lens 41a is imaged for each photoelectric element 42a, as in the case where a light-shielding wall 300 is placed between the photoelectric element 42a and the lens array 41 (see Figure 4) or where the photoelectric element 42a and the lens array 41 are in close contact (see Figure 5). 【0049】 Furthermore, the focal length of CIS4 is greater than the distance from the CIS4-side surface of the flat glass 201 to the CIS4-side surface of the pellicle 203, but less than 60mm (less than 40mm in this embodiment). Generally, the depth of field increases as the shooting distance increases, but since CIS4 does not have an aperture and its focal length is less than 60mm (less than 40mm in this embodiment), the depth of field of CIS4 is shallow. In this embodiment, the depth of field of CIS4 is smaller than the distance between the flat glass 201 and the mask 202, and between the mask 202 and the pellicle 203 (for example, smaller than half of each of these distances). 【0050】 The image data captured and generated by the CIS4 is converted into a digital signal within the CIS4 and then transferred in digital signal form to the control device 7 (image acquisition unit 74, described later). The control device 7 then stores the transferred image data and performs various image processing and calculation processes based on the image data. 【0051】 The scanning mechanism 5 is a mechanism that moves the stage 2 in the X-axis direction (left-right direction in Figure 1). By moving the stage 2, the mask set 200 is positioned relative to the CIS 4 along the X-axis direction. The scanning mechanism 5 may also be capable of moving the stage 2 in the Y-axis direction (front-back direction in Figure 1). 【0052】 The focusing position adjustment mechanism 6 adjusts the focusing position of the CIS4 to match the flat glass 201, mask 202, and pellicle 203 that are to be inspected. The focusing position adjustment mechanism 6 is configured to allow the CIS4 to move in the Z-axis direction (up and down direction in Figure 1), and by moving the CIS4, it is possible to individually focus on the circuit pattern surface of the mask 202, the upper surface of the flat glass 201, and the lower surface of the pellicle 203 (i.e., each of the surfaces to be inspected). 【0053】 The control device 7 performs various controls, image processing, and calculation processing in the mask set inspection device 1, including drive control of the CIS4 and the focus position adjustment mechanism 6. The control device 7 consists of a computer including a CPU (Central Processing Unit) that performs predetermined calculation processing, a ROM (Read Only Memory) that stores various programs and fixed value data, a RAM (Random Access Memory) that temporarily stores various data when various calculation processing is performed, and peripheral circuits for these. 【0054】 The control device 7 functions as various functional units, such as the main control unit 71, illumination control unit 72, CIS control unit 73, image acquisition unit 74, data processing unit 75, scanning control unit 76, focus position control unit 77, learning unit 78, and inspection unit 79, as the CPU operates according to various programs. 【0055】 However, the various functional units described above are realized through the cooperation of various hardware components such as the CPU, ROM, and RAM, and there is no need to clearly distinguish between functions realized in hardware and functions realized in software. Some or all of these functions may be realized by hardware circuits such as ICs. 【0056】 Furthermore, the control device 7 includes an input unit 7a consisting of a keyboard, mouse, touch panel, etc., a display unit 7b equipped with a display screen consisting of a liquid crystal display, etc., a storage unit 7c capable of storing various data, programs, calculation results, inspection results, etc., and a communication unit 7d capable of sending and receiving various data with the outside. 【0057】 Here, the various functional units that constitute the control device 7 will be described in detail. 【0058】 The main control unit 71 is a functional unit that controls the entire mask set inspection device 1 and is configured to send and receive various signals with other functional units such as the lighting control unit 72 and the CIS control unit 73. 【0059】 The lighting control unit 72 is a functional unit that drives and controls the lighting device 3, and performs switching control related to the on / off of the illuminated light based on command signals from the main control unit 71. 【0060】 The CIS control unit 73 is a functional unit that drives and controls the CIS 4, and controls the imaging timing and other parameters based on command signals from the main control unit 71. 【0061】 The image acquisition unit 74 is a functional unit for acquiring image data captured and acquired by the CIS4. 【0062】 The data processing unit 75 is a functional unit that performs predetermined image processing on image data acquired by the image acquisition unit 74, and performs measurement processing using the image data. 【0063】 The scanning control unit 76 is a functional unit that drives and controls the scanning mechanism 5, and controls the position of the scanning mechanism 5 based on command signals from the main control unit 71. In this embodiment, the scanning control unit 76 drives and controls the scanning mechanism 5, thereby moving any inspection area of ​​the fixed mask set 200 vertically downward from the CIS 4. Then, while sequentially moving the multiple inspection areas set on the mask 202, the flat glass 201, and the pellicle 203 vertically downward from the CIS 4, the inspection of each inspection area is performed, thereby completing the inspection of all inspection areas. 【0064】 The focus position control unit 77 is a functional unit that drives and controls the focus position adjustment mechanism 6, and controls the position of the CIS 4 based on command signals from the main control unit 71. In this embodiment, the focus position control unit 77 can adjust the focus position of the CIS 4 to a position where it is in focus with respect to the object to be inspected (flat glass 201, mask 202, or pellicle 203) by driving and controlling the focus position adjustment mechanism 6. 【0065】 The learning unit 78 is a functional unit that uses training data to train the deep neural network 90 (hereinafter simply referred to as "neural network 90"; see Figure 6) and constructs the first AI (Artificial Intelligence) model 101 as a "discrimination means for flat glass," the second AI model 102 as a "discrimination means for masks," and the third AI model 103 as a "discrimination means for pellicles." 【0066】 Each AI model 101, 102, and 103 in this embodiment is a generative model constructed by deep learning a neural network 90 using only image data related to an inspection target (flat glass 201, mask 202, or pellicle 203) free of foreign matter, as will be described later, and has the structure of a so-called autoencoder. 【0067】 Here, the structure of the neural network 90 will be explained with reference to Figure 6. Figure 6 is a schematic diagram conceptually showing the structure of the neural network 90. ​​As shown in Figure 6, the neural network 90 has the structure of a convolutional auto-encoder (CAE), comprising an encoder unit 91 as an "encoding unit" that extracts feature quantities (latent variables) TA from the input image data GA, and a decoder unit 92 as a "decoding unit" that reconstructs image data GB from the feature quantities TA. 【0068】 The structure of the convolutional autoencoder is well known, so a detailed explanation will be omitted. The encoder unit 91 has multiple convolutional layers 93, and in each convolutional layer 93, the result of a convolution operation using multiple filters (kernels) 94 on the input data is output as input data for the next layer. Similarly, the decoder unit 92 has multiple deconvolutional layers 95, and in each deconvolutional layer 95, the result of a deconvolution operation using multiple filters (kernels) 96 on the input data is output as input data for the next layer. Then, in the learning process described later, the weights (parameters) of each filter 94, 96 are updated. 【0069】 The inspection unit 79 is a functional unit that inspects whether or not foreign matter is attached to the object to be inspected (flat glass 201, mask 202, and pellicle 203). In this embodiment, the inspection unit 79 constitutes the "determination means". 【0070】 The memory unit 7c is composed of an HDD (Hard Disk Drive) or SSD (Solid State Drive), and has a predetermined memory area for storing, for example, each AI model 101, 102, 103 (neural network 90 and the learning information acquired through its learning). 【0071】 The communication unit 7d is equipped with a wireless communication interface conforming to communication standards such as wired LAN (Local Area Network) or wireless LAN, and is configured to send and receive various data with the outside. For example, the results of inspections performed by the inspection unit 79 are output to the outside via the communication unit 7d. 【0072】 Next, the learning process of the neural network 90 performed by the mask set inspection device 1 will be explained with reference to the flowchart in Figure 7. First, prior to the learning process, a large number of image data related to the inspection target (flat glass 201, mask 202, or pellicle 203) free of foreign matter are prepared as training source image data Ig1, Ig2, and Ig3 (see Figures 8-10; hereafter, these may be abbreviated as "training source image data Ig1-3"). Training source image data Ig1 is image data related to the flat glass 201, training source image data Ig2 is image data related to the mask 202, and training source image data Ig3 is image data related to the pellicle 203. 【0073】 In this embodiment, the training source image data Ig1 to Ig3 are prepared by actually focusing the CIS4 on an inspection target (flat glass 201, mask 202, or pellicle 203) that is free of foreign matter and imaging the said inspection target. Because the depth of field of the CIS4 is shallow, the circuit pattern of the mask 202 appears blurred in the training source image data Ig1 and Ig3 (in Figure 8, etc., the blurred state is represented by a dotted line). On the other hand, in the training source image data Ig2, the circuit pattern of the mask 202 appears clearly. 【0074】 Furthermore, instead of using image data Ig1-3 as training source image data, it is also possible to use virtually generated images (virtual image data) of an object free of foreign objects. Additionally, training source image data Ig1-3 may be image data obtained by CIS4 without any special processing (for example, monochrome luminance image data), or image data obtained by applying predetermined processing to the image data obtained by CIS4. 【0075】 Next, the first training data G1 is created from the training source image data Ig1, the second training data G2 from the training source image data Ig2, and the third training data G3 from the training source image data Ig3 (see Figures 11-13). In this embodiment, the first training data G1 is used to generate the first AI model 101, the second training data G2 is used to generate the second AI model 102, and the third training data G3 is used to generate the third AI model 103. 【0076】 Training data G1, G2, and G3 (hereinafter sometimes abbreviated as "training data G1-3") are obtained by pasting the training source image data Ig1-3 onto a predetermined image frame W. If a sufficient number of training source image data Ig1-3 cannot be prepared, training data G1-3 may be obtained by moving the training source image data Ig1-3 in the X or Y direction, or by rotating them by a predetermined angle, before pasting them onto the image frame W. 【0077】 Then, by using multiple training source image data Ig1-3, the required number of training data G1-3 are ultimately obtained. 【0078】 Once the required number of training data G1-3 have been acquired, in step S101, the learning unit 78 prepares an untrained neural network 90 based on a command from the main control unit 71. For example, it reads a neural network 90 that has been previously stored in the memory unit 7c or the like. Alternatively, it constructs a neural network 90 based on network configuration information (for example, the number of layers in the neural network and the number of nodes in each layer) stored in the memory unit 7c or the like. 【0079】 In this embodiment, three separate neural networks 90 are constructed: one for training using the first training data G1, one for training using the second training data G2, and one for training using the third training data G3. Therefore, in this embodiment, a total of three neural networks 90 are constructed. 【0080】 In the following step S102, reconstructed image data is acquired. Specifically, based on a command from the main control unit 71, the learning unit 78 provides the learning data G1 to 3 as input data to the input layer of the neural network 90, thereby acquiring the reconstructed image data output from the output layer of the neural network 90. ​​More specifically, the learning unit 78 provides the input layer of the neural network 90 with the learning data G1 to 3 that corresponds to the neural network 90 as input data, thereby acquiring the reconstructed image data output from the output layer of the neural network 90. ​​For example, the learning unit 78 provides the first learning data G1 as input data to the input layer of the neural network 90 that performs learning using the first learning data G1, and acquires the reconstructed image data output from the neural network 90. ​​In other words, the learning unit 78 inputs the appropriate learning data G1 to 3 to each of the three types of neural networks 90, and acquires the output reconstructed image data. 【0081】 In the following step S103, the learning unit 78 compares the input learning data G1-3 with the reconstructed image data output by the neural network 90 and determines whether the error is sufficiently small (whether it is below a predetermined threshold). 【0082】 Here, if the error is sufficiently small, in step S105, the learning unit 78 determines whether the learning termination conditions are met. For example, if a certain number of consecutive affirmative judgments are made in step S103 without going through the process of step S104 described later, or if learning using all of the prepared learning data G1 to G3 is repeated a predetermined number of times, it is determined that the termination conditions are met. If the termination conditions are met, the neural network 90 and its learning information (updated parameters, etc., described later) are stored in the memory unit 7c as AI models 101 to 103, and this learning process is terminated. 【0083】 In this embodiment, ultimately, a first AI model 101 for inspecting the flat glass 201, which is trained on first training data G1 related to the flat glass 201, a second AI model 102 for inspecting the mask 202, which is trained on second training data G2 related to the mask 202, and a third AI model 103 for inspecting the pellicle 203, which is trained on third training data G3 related to the pellicle 203, are stored in the memory unit 7c. Thus, a total of three types of AI models are stored. 【0084】 On the other hand, if the termination condition is not met in step S105, the process returns to step S101 and the neural network 90 is trained again. 【0085】 Furthermore, if the error is not sufficiently small in step S103, the network update process (training of the neural network 90) is performed in step S104, and then the process returns to step S102, repeating the above series of processes. 【0086】 Specifically, in the network update process of step S104, known learning algorithms such as backpropagation are used to update the weights (parameters) of each filter 94 and 96 in the neural network 90 to more appropriate values ​​so that the loss function representing the difference between the training data G1-3 and the reconstructed image data is minimized. For example, BCE (Binary Cross-entropy) can be used as the loss function. 【0087】 By repeatedly performing steps S102-S104, the neural network 90 minimizes the error between the training data G1-S3 and the reconstructed image data, resulting in the output of more accurate reconstructed image data. 【0088】 The final AI models 101 to 103 will generate reconstructed image data that closely matches the image data when image data relating to an object free of foreign matter is input. For example, the first AI model 101 will generate reconstructed image data that closely matches the image data relating to a flat glass 201 free of foreign matter is input. 【0089】 On the other hand, when each AI model 101 to 103 receives image data relating to an inspection object containing foreign matter, it generates reconstructed image data that closely matches the image data after the noise portion (the portion corresponding to the foreign matter) has been removed. For example, when the first AI model 101 receives image data relating to a flat glass 201 containing foreign matter, it generates reconstructed image data that closely matches the image data after the foreign matter has been removed. In other words, when a foreign matter is attached to an inspection object, a virtual image data relating to the inspection object is generated as the reconstructed image data relating to the inspection object, assuming that there is no foreign matter. 【0090】 Furthermore, it is not necessary to perform the above learning process each time the control device 7 is manufactured. The neural network 90 and its learning information (updated parameters, etc.) may be acquired in advance and stored in the memory unit 7c of the control device 7 as AI models 101 to 103. 【0091】 Next, the inspection process performed by the mask set inspection device 1 will be explained with reference to the flowchart in Figure 14. This inspection process is performed for each area to be inspected in the object to be inspected (mask 202, flat glass 201, or pellicle 203). In this embodiment, the inspection is performed first in the order of flat glass 201, mask 202, and pellicle 203. 【0092】 When the inspection process begins, in step S301, based on a command from the main control unit 71, the focus position control unit 77 moves the CIS4 to a position where it is in focus on the upper surface of the flat glass 201 (i.e., the surface to be inspected). 【0093】 Next, in step S302, the raw image data acquisition process is performed. In the raw image data acquisition process, the raw inspection image data Ik1 (see Figures 15 and 16) relating to the flat glass 201 to be inspected is acquired. In the raw image data acquisition process, imaging is performed by the CIS 4 while the position of the mask set 200 is gradually changed by the scanning mechanism 5, thereby acquiring the raw inspection image data Ik1 relating to at least one inspection area. The raw inspection image data Ik1 is the image data used to obtain the inspection image data K1 described later. The acquired raw inspection image data Ik1 is stored in the storage unit 7c. 【0094】 Furthermore, the original image data Ik1 for inspection may be image data obtained by CIS4 without any special processing (for example, monochrome luminance image data), or it may be image data obtained by applying predetermined processing to the image data obtained by CIS4 (the same applies to the original image data Ik2 and Ik3 for inspection described later). 【0095】 Furthermore, Figure 15 shows an example of the raw inspection image data Ik1 for a flat glass 201 without foreign matter, and Figure 16 shows an example of the raw inspection image data Ik1 for a flat glass 201 with foreign matter X1 (e.g., lint) attached to it. Due to the shallow depth of field of the CIS4, in the raw inspection image data Ik1, the circuit pattern of the mask 202 and foreign matter attached to the mask 202 and pellicle 203 are blurred, while the foreign matter X1 attached to the flat glass 201 is clearly visible. 【0096】 Next, in step S303, the inspection image data acquisition process is executed. In the inspection image data acquisition process, inspection image data K1 (see Figures 17 and 18) is acquired based on the original inspection image data Ik1 obtained in the original image data acquisition process. That is, inspection image data K1 is acquired by pasting the original inspection image data Ik1 onto the image frame W. Note that the image frame W used in the inspection process and the image frame W used in the learning process are the same. 【0097】 In the following step S304, the reconstructed image data acquisition process is performed. Specifically, based on a command from the main control unit 71, the inspection unit 79 inputs the inspection image data K1 acquired in step S303 into the input layer of the first AI model 101 corresponding to the type of inspection image data K1. The image data reconstructed by the first AI model 101 and output from the output layer is then acquired as the first reconstructed image data S1 (see Figure 19). The acquired first reconstructed image data S1 is stored in association with the inspection image data K1 from which the image data S1 was derived. 【0098】 Here, when the first AI model 101 receives inspection image data K1 (see Figure 18) relating to a flat glass 201 with foreign matter X1 attached, it outputs, as learned as described above, image data S1 relating to a flat glass 201 free of foreign matter, from which the foreign matter X1 has been removed, as the first reconstructed image data (see Figure 19). 【0099】 On the other hand, when the first AI model 101 receives inspection image data K1 (see Figure 17) relating to a flat glass 201 free of foreign matter, it outputs image data S1 relating to a flat glass 201 free of foreign matter that is almost identical to the inspection image data K1 (see Figure 19). The size (width and height) of the reconstructed image data S1 is the same as the size of the original inspection image data K1. 【0100】 In step S305, a pass / fail judgment process is performed based on the acquired first reconstructed image data S1. In the pass / fail judgment process, based on a command from the main control unit 71, the inspection unit 79 compares the entire inspection image data K1 acquired in step S303 with the entire first reconstructed image data S1 acquired in step S304 using the inspection image data K1, and calculates the difference between the two image data K1 and S1. For example, the dots (pixels) at the same coordinates in both image data K1 and S1 are compared, and the area (number of dots) of clusters of dots where the difference in brightness is greater than or equal to a predetermined value is calculated. Alternatively, instead of area, the length of the foreign object or the length of the longer side of a rectangle circumscribing the foreign object may be calculated as the difference. 【0101】 Next, the inspection unit 79 determines whether the calculated difference is greater than a predetermined threshold. If the calculated difference is greater than the predetermined threshold, the inspection unit 79 determines that "foreign matter is present." On the other hand, if the difference is less than the predetermined threshold, the inspection unit 79 determines that "no foreign matter is present." 【0102】 Furthermore, the inspection unit 79 performs the above determination for all inspection image data K1 relating to all areas to be inspected on the flat glass 201. If it determines that there are "no foreign objects" for all inspection image data K1, it determines that there are "no foreign objects" for the flat glass 201 and stores this result in the storage unit 7c. On the other hand, if, as a result of performing the above determination for all inspection image data K1, the inspection unit 79 determines that there are "foreign objects" for at least one inspection image data K1, it determines that there are "foreign objects" for the flat glass 201 and stores this result in the storage unit 7c. 【0103】 After the inspection process for the flat glass 201, the inspection unit 79 performs the same inspection process on the mask 202 as described above. 【0104】 In the inspection process targeting the mask 202, first, in step S301, the CIS4 is moved by the focus position adjustment mechanism 6 to a position where it is in focus on the circuit pattern surface of the mask 202 (i.e., the surface to be inspected). Next, in step S302, the raw inspection image data Ik2 (see Figures 20 and 21) related to the mask 202 to be inspected is acquired. Figure 20 shows an example of the raw inspection image data Ik2 for a mask 202 without foreign matter, and Figure 21 shows an example of the raw inspection image data Ik2 for a mask 202 with foreign matter X2 (e.g., lint) attached to it. Because the depth of field of the CIS4 is shallow, in the raw inspection image data Ik2, foreign matter attached to the flat glass 201 and pellicle 203 is blurred, while the circuit pattern of the mask 202 and the foreign matter X2 attached to the mask 202 are clearly visible. 【0105】 Next, in step S303, the original inspection image data Ik2 is pasted onto the image frame W to obtain inspection image data K2 (see Figures 22 and 23). Then, in step S304, the inspection image data K2 is input to the input layer of the second AI model 102 corresponding to the type of inspection image data K2. The image data reconstructed by the second AI model 102 and output from the output layer is then obtained as the second reconstructed image data S2 (see Figure 24). 【0106】 Here, when the second AI model 102 receives inspection image data K2 (see Figure 23) relating to the mask 202 with the foreign object X2 attached, it outputs, as learned as described above, image data S2 relating to the mask 202 with the foreign object X2 removed and free of foreign objects (see Figure 24). 【0107】 On the other hand, when the second AI model 102 receives inspection image data K2 (see Figure 22) relating to a mask 202 free of foreign matter, it outputs image data S2 relating to a mask 202 free of foreign matter that is almost identical to the inspection image data K2 (see Figure 24). 【0108】 Furthermore, in step S305, the inspection unit 79 compares the entire inspection image data K2 with the entire second reconstructed image data S2 and calculates the difference between the two image data K2 and S2. The inspection unit 79 then determines that "foreign matter is present" if the calculated difference is greater than a predetermined threshold, and determines that "no foreign matter is present" if the difference is less than a predetermined threshold. 【0109】 Furthermore, the inspection unit 79 performs the above determination for all inspection image data K2 relating to all areas to be inspected in the mask 202. If it determines that there are "no foreign objects" for all inspection image data K2, it determines that there are "no foreign objects" for the mask 202 and stores this result in the storage unit 7c. On the other hand, if, as a result of performing the above determination for all inspection image data K2, the inspection unit 79 determines that there are "foreign objects" for at least one inspection image data K2, it determines that there are "foreign objects" for the mask 202 and stores this result in the storage unit 7c. 【0110】 After the inspection process for mask 202, the inspection unit 79 performs the same inspection process for pellicle 203 as described above. 【0111】 In the inspection process targeting the pellicle 203, first, in step S301, the CIS4 is moved to a position where it is in focus on the lower surface of the pellicle 203 (i.e., the surface to be inspected), and in step S302, the raw inspection image data Ik3 (see Figures 25 and 26) related to the pellicle 203 to be inspected is acquired. Figure 25 shows an example of the raw inspection image data Ik3 related to a pellicle 203 without foreign matter, and Figure 26 shows an example of the raw inspection image data Ik3 related to a pellicle 203 with foreign matter X3 (e.g., lint) attached. Due to the shallow depth of field of the CIS4, in the raw inspection image data Ik3, the circuit pattern of the mask 202 and foreign matter attached to the mask 202 and the flat glass 201 are blurred, while the foreign matter X3 attached to the pellicle 203 is clearly visible. 【0112】 Next, in step S303, the original inspection image data Ik3 is pasted onto the image frame W to obtain inspection image data K3 (see Figures 27 and 28). Then, in step S304, the inspection image data K3 is input to the input layer of the third AI model 103, and the image data reconstructed by the third AI model 103 and output from the output layer is obtained as the third reconstructed image data S3 (see Figure 29). 【0113】 Here, when the third AI model 103 receives inspection image data K3 (see Figure 28) relating to the pellicle 203 with the foreign object X3 attached, it outputs, as learned in the above manner, image data S3 relating to the pellicle 203 without the foreign object X3, as the third reconstructed image data (see Figure 29). 【0114】 On the other hand, when the third AI model 103 receives inspection image data K3 (see Figure 28) relating to a pellicle 203 free of foreign matter, it outputs image data S3 relating to a pellicle 203 free of foreign matter that is almost identical to the inspection image data K3 (see Figure 29). 【0115】 Furthermore, in step S305, the inspection unit 79 compares the entire inspection image data K3 with the entire third reconstructed image data S3, and calculates the difference between the two image data sets K3 and S3. If the calculated difference is greater than a predetermined threshold, it is determined that "foreign matter is present," while if the difference is less than the predetermined threshold, it is determined that "no foreign matter is present." 【0116】 Furthermore, the inspection unit 79 performs the above determination for all inspection image data K3 relating to all areas to be inspected on the pellicle 203. If it determines that there are "no foreign objects" for all inspection image data K3, it determines that there are "no foreign objects" for the pellicle 203 and stores this result in the storage unit 7c. On the other hand, if, as a result of performing the above determination for all inspection image data K3, the inspection unit 79 determines that there are "foreign objects" for at least one inspection image data K3, it determines that there are "foreign objects" for the pellicle 203 and stores this result in the storage unit 7c. 【0117】 Then, the mask set inspection device 1 performs the above inspection process on all areas to be inspected in the flat glass 201, mask 202, and pellicle 203. If it determines that there are "no foreign matter" in all areas to be inspected, it determines that the mask set 200 is free of foreign matter (pass judgment) and stores this result in the storage unit 7c. 【0118】 On the other hand, if the mask set inspection device 1 finds that even one area under inspection is found to have foreign matter attached, it determines that the mask set 200 has foreign matter attached (failure judgment), stores this result in the storage unit 7c, and notifies the external party of the judgment result and the inspection target to which the foreign matter was attached via the display unit 7b, communication unit 7d, etc. 【0119】 As detailed above, according to this embodiment, in the image data obtained by CIS4, the circuit pattern of the mask 202 and foreign matter attached to inspection targets other than the target inspection target are blurred, while foreign matter attached to the target inspection target is clearly visible. Therefore, the presence or absence of foreign matter attached to the mask 202, the flat glass 201, and the pellicle 203 can be inspected individually and accurately without separating them. Furthermore, since it is not necessary to move the inspection targets individually during inspection, contamination of the mask 202 and other components due to movement can be prevented more reliably. 【0120】 Furthermore, by setting the focal length of CIS4 to less than 40mm, the circuit pattern of the mask 202 and foreign matter attached to objects other than the target object become more blurred in the image data obtained by CIS4, while foreign matter attached to the target object becomes more clearly visible. Therefore, the presence or absence of foreign matter can be inspected with greater accuracy. 【0121】 In addition, the CIS4 captures light that has been irradiated by the illumination device 3 and transmitted through the mask set 200. Therefore, the CIS4 can obtain image data that simulates the transfer of the circuit pattern (image data showing what will be transferred to the substrate). This makes it possible to more appropriately detect foreign objects that affect the transfer (exposure) of the circuit pattern. 【0122】 Furthermore, the focal length of CIS4 is set to be greater than the distance from the CIS4-side surface on the flat glass 201 to the CIS4-side surface on the pellicle 203. Therefore, when adjusting the focusing position of CIS4 to match the object being inspected, it is possible to focus on the object being inspected without bringing CIS4 into contact with the flat glass 201. This prevents damage to the flat glass 201 while enabling accurate inspection for the presence or absence of foreign matter on each object being inspected. 【0123】 Furthermore, the lighting device 3 can be placed on the pellicle 203 side and the CIS 4 on the flat glass 201 side, and it is not necessary to provide the lighting device 3 and CIS 4 on both the flat glass 201 side and the pellicle 203 side, respectively. Therefore, it is possible to suppress increases in manufacturing and maintenance costs for the mask set inspection device 1 and to miniaturize and simplify the mask set inspection device 1. In particular, in this embodiment, since only one CIS 4 is provided, it is possible to suppress cost increases and miniaturize and simplify the mask set inspection device 1 more effectively. 【0124】 In addition, the illumination device 3 irradiates the mask set 200 with diffused ultraviolet light with a wavelength of 365 nm. Therefore, the mask set 200 can be irradiated with light similar to the light (so-called i-line) used during the transfer (exposure) of the circuit pattern by the exposure device. As a result, the CIS 4 can more reliably obtain image data that simulates the transfer of the circuit pattern. Consequently, foreign substances that affect the transfer (exposure) of the circuit pattern can be detected more appropriately. 【0125】 Furthermore, in this embodiment, instead of light passing through one cylindrical lens 41a forming an image on one photoelectric conversion element 42a, light passing through multiple cylindrical lenses 41a forms an image on one photoelectric conversion element 42a. Therefore, it is possible to more reliably prevent foreign matter from being hidden by circuit patterns, etc., in the image data obtained by CIS4. This makes it possible to inspect for the presence or absence of foreign matter with even greater accuracy. 【0126】 In addition, in this embodiment, the inspection image data K1, K3, K3 obtained by CIS4 relating to the target inspection object are compared with the reconstructed image data S1, S2, S3 which are reconstructed by inputting the image data K1, K2, K3 into AI models 101-103. Based on the comparison result, the presence or absence of foreign matter in the target inspection object is determined. Since both image data to be compared relate to the same inspection object, foreign matter inspection can be performed more easily and accurately. 【0127】 Furthermore, since the shape and appearance of the mask set 200 (e.g., the shape and position of the circuit pattern) are almost identical in both image data being compared, unlike methods that detect foreign objects by comparing them to a reference image, it is not necessary to set relatively lenient inspection conditions to prevent false detections, and stricter inspection conditions can be set. In addition, the state of the object being inspected (e.g., the placement position, placement angle, and deflection of the mask set 200) and imaging conditions (e.g., lighting conditions and the field of view of the CIS4) can be matched in both image data being compared. These combined effects allow for even more accurate inspection of whether or not foreign objects are present. 【0128】 Furthermore, the embodiment is not limited to the description above, and may be implemented as follows, for example. Of course, other applications and modifications not exemplified below are also possible. 【0129】 (a) In the above embodiment, one CIS4 is provided, and this CIS4 is configured to image the mask 202, the flat glass 201, and the pellicle 203, respectively. In contrast, as shown in Figure 30, a total of three CISs may be provided: a first CIS81 for imagering the flat glass 201, a second CIS82 for imagering the mask 202, and a third CIS83 for imagering the pellicle 203. In this case, the focus position can be adjusted and imaging can be performed simultaneously in each CIS81,82,83, so image data related to each inspection target can be acquired more efficiently compared to when each inspection target is imaged with one CIS4. Therefore, the cycle time for inspection can be effectively shortened. In this example, the first CIS81 constitutes the "first imaging means," the second CIS82 constitutes the "second imaging means," and the third CIS83 constitutes the "third imaging means." 【0130】 Alternatively, the focusing position adjustment mechanism 6 may include an overall adjustment mechanism 61 that can move all of the CIS 81, 82, and 83 at once, and individual adjustment mechanisms 62a, 62b, and 62c that can move the CIS 81, 82, and 83 individually. In this case, the CIS 81 and 82 are shifted along the Z-axis direction (the direction of movement of the CIS 82, etc.) by the same distance as the distance between the upper surface of the flat glass 201 and the circuit pattern surface of the mask 202, and the CIS 82 and 83 are shifted along the Z-axis direction by the same distance as the distance between the circuit pattern surface of the mask 202 and the lower surface of the pellicle 203. By moving all of the CIS 81, 82, and 83 with the overall adjustment mechanism 61, the focus of all of the CIS 81, 82, and 83 can be adjusted to the image target at once. Furthermore, the individual adjustment mechanisms 62a, 62b, and 62c allow for adjustment of the relative positions of CIS 81, 82, and 83 to match the configuration of the mask set 200 being inspected, as well as for fine-tuning during focusing. 【0131】 (b) In the above embodiment, the presence or absence of foreign matter is determined using AI models 101 to 103, but the presence or absence of foreign matter may be determined without using AI models. For example, foreign matter that interferes with the transfer (exposure) of the circuit pattern will have low brightness, so the presence or absence of foreign matter may be determined by brightness. Also, since foreign matter that is in focus will have a clear outer edge, the presence or absence of foreign matter may be determined by differentiation (for example, the rate of change in brightness). 【0132】 (c) In the above embodiment, the focus position adjustment mechanism 6 is configured to adjust the focus position of the CIS4 by moving the CIS4, but the focus position may be adjusted by adjusting other elements (for example, the lens array 41 or the optical path). 【0133】 (d) In the above embodiment, the illumination device 3 is configured to emit ultraviolet light with a wavelength of 365 nm, but the wavelength of the light emitted from the illumination device 3 may be changed as appropriate. However, in terms of improving inspection accuracy, it is preferable that the light emitted from the illumination device 3 has the same wavelength as the light used for transferring (exposing) the circuit pattern of the mask 202. 【0134】 (e) In the above embodiment, the CIS4 is configured such that light passing through multiple cylindrical lenses 41a is imaged onto one photoelectric conversion element 42a. Alternatively, a CIS4 may be employed in which only light passing through one cylindrical lens 41a is imaged onto one photoelectric conversion element 42a. [Explanation of symbols] 【0135】 1...Mask set inspection device, 3...Illumination device (irradiation means), 4...CIS (imaging means), 6...Focus position adjustment mechanism (focus position adjustment means), 31...Light source, 32...Diffuser plate, 41...Lens array, 41a...Cylindrical lens, 42...Photoelectric conversion element array, 42a...Photoelectric conversion element, 79...Inspection unit (determination means), 81...First CIS (first imaging means), 82...Second CIS (second imaging means), 83...Third CIS (third imaging means), 91...Encoder unit (encoding unit), 92...Decoder unit (decoding unit), 101...First AI model (identification means for flat glass), 102...Second AI model (identification means for mask), 103...Third AI model (identification means for pellicle), 200...Mask set, 201...Flat glass, 202...Mask, 203...Pellicle.

Claims

[Claim 1] A mask set inspection apparatus for inspecting for the adhesion of foreign matter to a mask set having a mask on which a circuit pattern for transfer onto a substrate is formed, and a transparent flat glass and a pellicle, respectively, arranged so as to sandwich the mask at a predetermined distance from the mask, the apparatus comprising: An irradiation means for irradiating the mask set with a predetermined light, The system includes a lens array formed by arranging multiple cylindrical lenses in parallel, and a photoelectric element array formed by arranging multiple photoelectric elements that image light passing through the cylindrical lenses, and an imaging means provided at a position sandwiching the mask set between the irradiation means and the mask set, capable of imaging light irradiated by the irradiation means and transmitted through the mask set. The mask, the flat glass, and the pellicle are each to be inspected, and the focusing position adjustment means is capable of adjusting the focusing position of the imaging means to match the inspected object. The system includes a determination means capable of determining the presence or absence of foreign matter in the object to be inspected based on image data obtained by the imaging means, The mask set inspection apparatus is characterized in that the imaging means has a focal length greater than the distance from the surface of the flat glass on the side of the imaging means to the surface of the pellicle on the side of the imaging means, but less than 60 mm. [Claim 2] The mask set inspection apparatus according to claim 1, characterized in that the imaging means has a focal length of less than 40 mm. [Claim 3] Only one imaging device is provided. The mask set inspection apparatus according to claim 1, characterized in that, for each of the inspection targets, the focus position of the imaging means is adjusted by the focus position adjustment means, and imaging is performed by the imaging means, and the presence or absence of foreign matter is determined by the determination means based on the image data obtained by this imaging. [Claim 4] The irradiation means is A light source that emits ultraviolet light with a wavelength of 365 nm, The mask set inspection apparatus according to claim 1, further comprising a diffuser plate that diffuses ultraviolet light emitted from the light source and irradiates the mask set with it. [Claim 5] The mask set inspection apparatus according to claim 1, characterized in that the imaging means is configured such that light passing through a plurality of cylindrical lenses forms an image on one of the photoelectric conversion elements. [Claim 6] A neural network having an encoding unit that extracts features from input image data and a decoding unit that reconstructs image data from the features is trained using only image data of the flat glass free of foreign objects to generate a flat glass identification means, A mask identification means generated by training the aforementioned neural network with only image data relating to the mask free of foreign objects as training data, The neural network is trained using only image data of the pellicle free of foreign objects as training data to generate a pellicle identification means, The determination means is, By comparing the image data relating to the flat glass obtained by the imaging means with the first reconstructed image data, which is image data reconstructed by inputting the image data into the identification means for flat glass, the presence or absence of foreign matter in the flat glass is determined. By comparing the image data relating to the mask obtained by the imaging means with the second reconstructed image data, which is image data reconstructed by inputting the image data into the mask identification means, the presence or absence of foreign matter in the mask is determined. The mask set inspection apparatus according to claim 1, characterized in that it is configured to determine the presence or absence of foreign matter on the pellicle by comparing image data relating to the pellicle obtained by the imaging means with a third reconstructed image data which is image data reconstructed by inputting the image data into the pellicle identification means. [Claim 7] The mask set inspection apparatus according to claim 1, characterized in that it comprises a total of three imaging means: a first imaging means for imaging the flat glass, a second imaging means for imaging the mask, and a third imaging means for imaging the pellicle.

Images